Many organizations have already raised their teams' awareness of artificial intelligence. Participants know the key concepts and have seen demonstrations. Yet, practices remain scattered, rules are poorly understood and difficult questions arise on a case-by-case basis.
Awareness is an entry point. An AI culture then requires real use cases, understandable rules, explicit responsibilities and feedback loops. Since 2 February 2025, Article 4 of the AI Act has required providers and deployers of AI systems to take measures to ensure a sufficient level of AI literacy, tailored to the people involved and their contexts of use.
An awareness session opens a conversation. It does not sustain it over time.
A well-designed conference can have several useful effects: provide reference points, correct misconceptions, put concerns into words and create a shared experience. It can also create a desire to try.
Its effect stops there if the organisation does not answer the questions that follow:
- which tool can I use?
- can I upload this document to it?
- who checks the answer?
- Should I declare that I used an AI?
- what should I do if the result seems wrong?
- how to share a useful practice?
Without answers, the initial momentum fragments. The most confident experiment alone. The most cautious wait. Rules circulate as rumours. The real culture develops despite the organisation, rather than with it.
Culture is not measured by what people have heard. It is visible in how they decide when a situation was not anticipated.
Article 4 of the AI Act requires AI literacy suited to the context.
Since 2 February 2025, Article 4 of the EU AI Act has applied. It requires providers and deployers of AI systems to take measures, to their best extent, to ensure a sufficient level of AI literacy among their staff and other people who operate these systems on their behalf.
The text requires taking into account:
- technical knowledge;
- experience;
- training;
- the context in which the system is used;
- people or groups on which it is used.
It prescribes neither a specific qualification, a fixed number of hours nor an identical programme for everyone. This flexibility does not reduce the requirement. It means training must be connected to actual practices.
The European Commission has published a directory of more than 40 practices. It includes e-learning, in-person training, bootcamps, and collaborations with academia. It specifies that reproducing a practice from the directory does not automatically create a presumption of compliance.
Therefore, the issue is not simply checking a box. It is being able to explain why the framework is suited to the people, tools, and risks of the organization.
A strong AI culture rests on four layers.
Shared reference points
Everyone must understand the general capabilities and limitations: probabilistic generation, plausible errors, sensitive data, intellectual property, bias and human accountability. Shared vocabulary makes dialogue possible.
Skills by role
Needs vary. A user must know how to frame and verify. A manager must arbitrate between practices and workload. The legal, security and purchasing functions must assess risks. Senior management must assign responsibilities.
Practical rules
A twenty-page charter that teams do not know offers little protection. Rules must be available at the point of action: approved tools, prohibited data, validation levels, people to contact and incident procedure.
A capacity for collective learning
Tools change, as do practices. The organization must collect feedback, document errors, update the rules, and share proven practices.
NIST’s framework translates this logic into continuous governance: training adapted to responsibilities, defined supervisory roles, documented usage boundaries and regular engagement from stakeholders.
Establish a few rituals that make culture visible.
These rituals avoid two extremes: the general prohibition that pushes uses into the shadows and unstructured experimentation that transfers all risks to individuals.
Move from awareness to an AI culture in 90 days.
| Period | Expected work | Evidence |
|---|---|---|
| Days 1 to 30 | Map tools, practices, roles, and main risk areas | A shared snapshot and priorities |
| Days 31 to 60 | Training by role and testing two or three defined use cases | Observable practices and documented feedback |
| Days 61 to 90 | Formalize the rules, responsibilities, indicators and rituals | A usable framework and an improvement loop |
The journey begins with a simple question: what must people be able to decide when they use AI in their work?
The answer provides a more precise program than a list of features. It links knowledge, practice, and responsibility.
Awareness remains valuable. It creates attention and a common language. But an AI culture is built when the organization transforms this attention into the ability to act, verify, discuss, and correct.
Reference sources
- EU Regulation 2024/1689, Article 4 on AI literacy
- European Commission, repository of AI literacy practices
- NIST, AI Risk Management Framework Core
- OECD, AI and skills, 2026
FAQ
What does AI literacy mean under the AI Act?
The regulation targets the skills, knowledge and understanding that enable informed deployment of AI systems and awareness of their possibilities, risks and potential harm. The expected level depends on individuals, their experience and the context of use.
Is a conference enough to comply with Article 4?
The regulation does not impose a single format. A conference can contribute to the framework, but an isolated and general measure may be insufficient if usage, risks and responsibilities differ significantly between teams.
Should everyone be trained in the same way?
No. An occasional user, a buyer, a business lead, a legal adviser and a technical team do not have the same tasks or responsibilities.
How do you know if an AI culture exists?
Teams know which tools to use, what data to protect, how to verify an output, when to ask for advice, how to report an incident, and who assumes responsibility for the decision.
